Sci Rep. 2025 Apr 28;15(1):14779. doi: 10.1038/s41598-025-99529-0.

ABSTRACT

Melanoma skin cancer is a deadly disease with a high mortality rate. A prompt diagnosis can aid in the treatment of the disease and potentially save the patient's life. Artificial intelligence methods can help diagnose cancer at a rapid speed. The literature has employed numerous Machine Learning (ML) and Deep Learning (DL) algorithms to detect skin cancer. ML algorithms perform well for small datasets but cannot comprehend larger ones. Conversely, DL algorithms exhibit strong performance on large datasets but misclassify when applied to smaller ones. We conduct extensive experiments using a convolutional neural network (CNN), varying its parameter values to determine which set of values yields the best performance measure. We discovered that adding layers, making each Conv2D layer have multiple filters, and getting rid of dropout layers greatly improves the accuracy of the classifiers, going from 62.5% to 85%. We have also discussed the parameters that have the potential to significantly impact the model's performance. This shows how powerful it is to fine-tune the parameters of a CNN-based model. These findings can assist researchers in fine-tuning their CNN-based models for use with skin cancer image datasets.

PMID:40295678 | DOI:10.1038/s41598-025-99529-0

Sci Rep. 2025 Apr 28;15(1):14851. doi: 10.1038/s41598-025-98724-3.

ABSTRACT

Addressing the challenges of integrating photovoltaic (PV) systems into power grids, this research develops a dual-phase optimization model incorporating deep learning techniques. Given the fluctuating nature of solar energy, the study employs Generative Adversarial Networks (GANs) to simulate diverse and high-resolution energy generation-consumption patterns. These synthetic scenarios are subsequently utilized within a real-time adaptive control framework, allowing for dynamic adjustments in operational strategies that enhance both efficiency and grid stability. By leveraging this approach, the model has demonstrated substantial improvements in economic and environmental performance, achieving up to 96% efficiency while reducing energy expenses by 20%, lowering carbon emissions by 30%, and cutting annual operational downtime by half (from 120 to 60 h). Through a scenario-driven predictive analysis, this framework provides data-driven optimization for energy systems, strengthening their resilience against renewable energy intermittency. Furthermore, the integration of AI-enhanced forecasting techniques ensures proactive decision-making, supporting a sustainable transition toward greener energy solutions.

PMID:40295668 | DOI:10.1038/s41598-025-98724-3

Sci Rep. 2025 Apr 28;15(1):14913. doi: 10.1038/s41598-025-98205-7.

ABSTRACT

Skin cancer represents a significant global public health issue, and prompt and precise detection is essential for effective treatment. This study introduces SkinEHDLF, an innovative deep-learning model that enhances skin cancer classification. SkinEHDLF utilizes the advantages of several advanced models, i.e., ConvNeXt, EfficientNetV2, and Swin Transformer, while integrating an adaptive attention-based feature fusion mechanism to enhance the synthesis of acquired features. This hybrid methodology combines ConvNeXt's proficient feature extraction capabilities, EfficientNetV2's scalability, and Swin Transformer's long-range attention mechanisms, resulting in a highly accurate and dependable model. The adaptive attention mechanism dynamically optimizes feature fusion, enabling the model to focus on the most relevant information, enhancing accuracy and reducing false positives. We trained and evaluated SkinEHDLF using the ISIC 2024 dataset, which comprises 401,059 skin lesion images extracted from 3D total-body photography. The dataset is divided into three categories: melanoma, benign lesions, and noncancerous skin anomalies. The findings indicate the superiority of SkinEHDLF compared to current models. In binary skin cancer classification, SkinEHDLF surpassed baseline models, achieving an AUROC of 99.8% and an accuracy of 98.76%. The model attained 98.6% accuracy, 97.9% precision, 97.3% recall, and 99.7% AUROC across all lesion categories in multi-class classification. SkinEHDLF demonstrates a 7.9% enhancement in accuracy and a 28% decrease in false positives, outperforming leading models including ResNet-50, EfficientNet-B3, ViT-B16, and hybrid methodologies such as ResNet-50 + EfficientNet and ViT + CNN, thereby positioning itself as a more precise and reliable solution for automated skin cancer detection. These findings underscore SkinEHDLF's capacity to transform dermatological diagnostics by providing a scalable and accurate method for classifying skin cancer.

PMID:40295588 | DOI:10.1038/s41598-025-98205-7

Comput Methods Programs Biomed. 2025 Apr 23;267:108801. doi: 10.1016/j.cmpb.2025.108801. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVE: Exploring the dependencies between multimodal brain networks and integrating node features to enhance brain disease diagnosis remains a significant challenge. Some work has examined only brain connectivity changes in patients, ignoring important information about radiomics features such as shape and texture of individual brain regions in structural images. To this end, this study proposed a novel deep learning approach to integrate multimodal brain connectome information and regional radiomics features for brain disease diagnosis.

METHODS: A dual-branch autoencoder (ConnectomeAE) based on multimodal brain connectomes was proposed for brain disease diagnosis. Specifically, a matrix of radiomics feature extracted from structural magnetic resonance image (MRI) was used as Rad_AE branch inputs for learning important brain region features. Functional brain network built from functional MRI image was used as inputs to Cycle_AE for capturing brain disease-related connections. By separately learning node features and connection features from multimodal brain networks, the method demonstrates strong adaptability in diagnosing different brain diseases.

RESULTS: ConnectomeAE was validated on two publicly available datasets. The experimental results show that ConnectomeAE achieved excellent diagnostic performance with an accuracy of 70.7 % for autism spectrum disorder and 90.5 % for Alzheimer's disease. A comparison of training time with other methods indicated that ConnectomeAE exhibits simplicity and efficiency suitable for clinical applications. Furthermore, the interpretability analysis of the model aligned with previous studies, further supporting the biological basis of ConnectomeAE.

CONCLUSIONS: ConnectomeAE could effectively leverage the complementary information between multimodal brain connectomes for brain disease diagnosis. By separately learning radiomic node features and connectivity features, ConnectomeAE demonstrated good adaptability to different brain disease classification tasks.

PMID:40294455 | DOI:10.1016/j.cmpb.2025.108801

J Med Internet Res. 2025 Apr 28;27:e58723. doi: 10.2196/58723.

ABSTRACT

In this viewpoint, we explore the use of big data analytics and artificial intelligence (AI) and discuss important challenges to their ethical, effective, and equitable use within opioid use disorder (OUD) treatment settings. Applying our collective experiences as OUD policy and treatment experts, we discuss 8 key challenges that OUD treatment services must contend with to make the most of these rapidly evolving technologies: data and algorithmic transparency, clinical validation, new practitioner-technology interfaces, capturing data relevant to improving patient care, understanding and responding to algorithmic outputs, obtaining informed patient consent, navigating mistrust, and addressing digital exclusion and bias. Through this paper, we hope to critically engage clinicians and policy makers on important ethical considerations, clinical implications, and implementation challenges involved in big data analytics and AI deployment in OUD treatment settings.

PMID:40294410 | DOI:10.2196/58723

PLoS One. 2025 Apr 28;20(4):e0321529. doi: 10.1371/journal.pone.0321529. eCollection 2025.

ABSTRACT

Accurate power load forecasting helps reduce energy waste and improve grid stability. This paper proposes a hybrid forecasting model, BiStacking+TCN-GRU, which leverages both ensemble learning and deep learning techniques. The model first applies the Pearson correlation coefficient (PCC) to select features highly correlated with the power load. Then, BiStacking is used for preliminary predictions, followed by a temporal convolutional network (TCN) enhanced by a gated recurrent unit (GRU) to produce the final predictions. The experimental validation based on Panama's 2020 electricity load data demonstrated the effectiveness of the model, with the model achieving an RMSE of 29.1213 and an MAE of 22.5206, respectively, with an R² of 0.9719. These results highlight the model's superior performance in short-term load forecasting, demonstrating its strong practical applicability and theoretical contributions.

PMID:40294011 | DOI:10.1371/journal.pone.0321529

IEEE Trans Med Imaging. 2025 Apr 28;PP. doi: 10.1109/TMI.2025.3565000. Online ahead of print.

ABSTRACT

Due to the difficulty of collecting multi-label annotations for retinal diseases, fundus images are usually annotated with only one label, while they actually have multiple labels. Given that deep learning requires accurate training data, incomplete disease information may lead to unsatisfactory classifiers and even misdiagnosis. To cope with these challenges, we propose a co-pseudo labeling and active selection method for Fundus Single-Positive multi-label learning, named FSP. FSP trains two networks simultaneously to generate pseudo labels through curriculum co-pseudo labeling and active sample selection. The curriculum co-pseudo labeling adjusts the thresholds according to the model's learning status of each class. Then, the active sample selection maintains confident positive predictions with more precise pseudo labels based on loss modeling. A detailed experimental evaluation is conducted on seven retinal datasets. Comparison experiments show the effectiveness of FSP and its superiority over previous methods. Downstream experiments are also presented to validate the proposed method.

PMID:40293917 | DOI:10.1109/TMI.2025.3565000

IEEE Trans Neural Netw Learn Syst. 2025 Apr 28;PP. doi: 10.1109/TNNLS.2025.3558596. Online ahead of print.

ABSTRACT

Accurate prediction of pathological subtypes on radiological images is one of the most important deep learning (DL) tasks for the appropriate selection of clinical treatment. It is challenging for conventional DL models to obtain sufficient pathological labels for training because of the heavy workload, invasive surgery, and knowledge requirements in pathological analysis. However, existing methods based on limited annotations, such as active learning (AL) and semi-supervised learning (SSL), have difficulty in capturing lesion's effective features because of the complicated semantic information of radiologic images. In this article, we introduce an efficient domain knowledge-guided semantic prediction framework that integrates domain knowledge-guided AL and SSL methods. This framework can effectively predict pathological subtypes on the basis of radiologic images with limited pathological annotations via three key modules: 1) the discriminative spatial-semantic feature extraction module captures the spatial-semantic features of lesions as semantic information that can better reflect the semantic relationship and effectively mitigate overfitting risk; 2) the explicit sign-guided anchor attention module measures the multimodal semantic distribution of samples under the guidance of clinical domain knowledge, thus selecting the most representative AL samples for pathological labeling; and 3) the implicit radiomics-guided dual-task entanglement module exploits the inherent constraint relationships between implicit radiomics features (IRFs) and pathological subtypes, facilitating the aggregation of unlabeled data. Experiments have been extensively conducted to evaluate our method in two clinical tasks: the pathological grading prediction in pancreatic neuroendocrine neoplasms (pNENs) and muscular invasiveness prediction in bladder cancer (BCa). The experimental results on both tasks demonstrate that the proposed method consistently outperforms the state-of-the-art approaches by a large margin.

PMID:40293902 | DOI:10.1109/TNNLS.2025.3558596

IEEE Trans Biomed Eng. 2025 Apr 28;PP. doi: 10.1109/TBME.2025.3561338. Online ahead of print.

ABSTRACT

Coronary CT angiography (CCTA) is an essential technique for clinical coronary assessment. However, the risks associated with ionizing radiation cannot be ignored, especially its stochastic effects, which increase the risk of cancer. Although it can effectively alleviate radiation problems, low-dose CCTA can reduce imaging quality and interfere with the diagnosis of the radiologist. Existing deep learning methods based on image restoration suffer from subtle structure degradation after noise suppression, leading to unclear coronary boundaries. Furthermore, in the absence of prior guidance on coronary location, subtle coronary branches will be lost after aggressive noise suppression and are difficult to restore successfully. To address the above issues, this paper proposes a novel Guided Refinement Network (GRN) model based on joint learning for restoring high-quality images from low-dose CCTA. GRN integrates coronary segmentation, which provides coronary location, into the denoising, and the two leverage mutual guidance for effective interaction and collaborative optimization. On the one hand, denoising provides images with lower noise levels for segmentation to assist in generating coronary masks. Furthermore, segmentation provides a prior coronary location for denoising, aiming to preserve and restore subtle coronary branches. GRN achieves noise suppression and subtle structure enhancement for low-dose CCTA imaging through joint denoising and segmentation, while also generating segmentation results with reference value. Quantitative and qualitative results show that GRN outperforms existing methods in noise suppression, subtle structure restoration, and visual perception improvement, and generates coronary masks that can serve as a reference for radiologists to assist in diagnosing coronary disease.

PMID:40293900 | DOI:10.1109/TBME.2025.3561338

IEEE J Biomed Health Inform. 2025 Apr 28;PP. doi: 10.1109/JBHI.2025.3561146. Online ahead of print.

ABSTRACT

Cerebrovascular segmentation is essential for diagnosing and treating cerebrovascular diseases. However, accurately segmenting cerebral vessels in TOF-MRA remains challenging due to significant interindividual variations in cerebrovascular morphology, low image con-trast, and class imbalance. The present study proposes an advanced deep learning model called PPA Net, consisting of VesselMRA Net and VesselConvLSTM components. Firstly, VesselMRA Net utilizes rectangular convolutional blocks to fuse multi-scale features, enhancing feature extraction per-formance. VesselMRA Net employs the attention mechanism to boost certain valuable semantic weighting, addressing segmentation challenges arising from class imbalance and low contrast. Secondly, VesselConvLSTM, a pixel-level prediction model, employs a gating mechanism to learn cerebral vessel morphology across individuals. It reduces individual differences in segmentation and restores inter-voxel correlations disrupted by data slicing, aiding VesselMRA Net in accurately segmenting cerebrovascular pixels. Lastly, integrating VesselMRA Net and VesselConv-LSTM results in a modular cerebral vessel segmentation framework, PPA Net, facilitating separate optimization of the backbone network and predicted model components. The performance of this model has been extensively validated through experimental evaluations on three publicly available datasets, obtaining significant competitiveness when compared to the state-of-the-art of the current cerebral vessel segmentation models.

PMID:40293899 | DOI:10.1109/JBHI.2025.3561146

IEEE J Biomed Health Inform. 2025 Apr 28;PP. doi: 10.1109/JBHI.2025.3564981. Online ahead of print.

ABSTRACT

Knee osteoarthritis (KOA) as a prevalent chronic disease, detrimentally impacts the quality of life among affected individuals. The knee adduction moment (KAM) during the stance phase has been identified as a potential biomechanical measure for assessing the severity of KOA. Traditional KAM assessment relies on expensive equipment, which limits its popularization. In contrast, current KAM estimation methods based on wearables and deep-learning technology offer the advantage of lower costs. However, it still suffers challenges in achieving accurate estimation. To address this challenge, a novel deep-learning framework is proposed in this work, which estimates the KAM from Inertial Measurement Units (IMU) and Electromyography (EMG) data by a well-designed self-aware fusion model. Walking data from 18 effective subjects were recorded with 4 IMUs and 6 EMGs. Results show that the model significantly improves KAM estimation accuracy. The relative root-mean-square error of the proposed model is 9.15% BW BH lower than counterpart estimation methods.

PMID:40293894 | DOI:10.1109/JBHI.2025.3564981

Abdom Radiol (NY). 2025 Apr 28. doi: 10.1007/s00261-025-04951-7. Online ahead of print.

ABSTRACT

Accurate, reproducible body composition analysis from abdominal computed tomography (CT) images is critical for both clinical research and patient care. We present a fully automated, artificial intelligence (AI)-based pipeline that streamlines the entire process-from data normalization and anatomical landmarking to automated tissue segmentation and quantitative biomarker extraction. Our methodology ensures standardized inputs and robust segmentation models to compute volumetric, density, and cross-sectional area metrics for a range of organs and tissues. Additionally, we capture selected DICOM header fields to enable downstream analysis of scan parameters and facilitate correction for acquisition-related variability. By emphasizing portability and compatibility across different scanner types, image protocols, and computational environments, we ensure broad applicability of our framework. This toolkit is the basis for the Opportunistic Screening Consortium in Abdominal Radiology (OSCAR) and has been shown to be robust and versatile, critical for large multi-center studies.

PMID:40293521 | DOI:10.1007/s00261-025-04951-7

Abdom Radiol (NY). 2025 Apr 28. doi: 10.1007/s00261-025-04963-3. Online ahead of print.

ABSTRACT

Renal cell carcinoma (RCC) as a significant health concern, with incidence rates rising annually due to increased use of cross-sectional imaging, leading to a higher detection of incidental renal lesions. Differentiation between benign and malignant renal lesions is essential for effective treatment planning and prognosis. Renal tumors present numerous histological subtypes with different prognoses, making precise subtype differentiation crucial. Artificial intelligence (AI), especially machine learning (ML) and deep learning (DL), shows promise in radiological analysis, providing advanced tools for renal lesion detection, segmentation, and classification to improve diagnosis and personalize treatment. Recent advancements in AI have demonstrated effectiveness in identifying renal lesions and predicting surveillance outcomes, yet limitations remain, including data variability, interpretability, and publication bias. In this review we explored the current role of AI in assessing kidney lesions, highlighting its potential in preoperative diagnosis and addressing existing challenges for clinical implementation.

PMID:40293518 | DOI:10.1007/s00261-025-04963-3

Int Orthop. 2025 Apr 28. doi: 10.1007/s00264-025-06531-2. Online ahead of print.

ABSTRACT

AIM: We aimed to compare the diagnostic performance of physicians in the detection of arthroscopically confirmed meniscus and anterior cruciate ligament (ACL) tears on knee magnetic resonance imaging (MRI), with and without assistance from a deep learning (DL) model.

METHODS: We obtained preoperative MR images from 88 knees of patients who underwent arthroscopic meniscal repair, with or without ACL reconstruction. Ninety-eight MR images of knees without signs of meniscus or ACL tears were obtained from a publicly available database after matching on age and ACL status (normal or torn), resulting in a global dataset of 186 MRI examinations. The Keros® (Incepto, Paris) DL algorithm, previously trained for the detection and characterization of meniscus and ACL tears, was used for MRI assessment. Magnetic resonance images were individually, and blindly annotated by three physicians and the DL algorithm. After three weeks, the three human raters repeated image assessment with model assistance, performed in a different order.

RESULTS: The Keros® algorithm achieved an area under the curve (AUC) of 0.96 (95% CI 0.93, 0.99), 0.91 (95% CI 0.85, 0.96), and 0.99 (95% CI 0.98, 0.997) in the detection of medial meniscus, lateral meniscus and ACL tears, respectively. With model assistance, physicians achieved higher sensitivity (91% vs. 83%, p = 0.04) and similar specificity (91% vs. 87%, p = 0.09) in the detection of medial meniscus tears. Regarding lateral meniscus tears, sensitivity and specificity were similar with/without model assistance. Regarding ACL tears, physicians achieved higher specificity when assisted by the algorithm (70% vs. 51%, p = 0.01) but similar sensitivity with/without model assistance (93% vs. 96%, p = 0.13).

CONCLUSIONS: The current model consistently helped physicians in the detection of medial meniscus and ACL tears, notably when they were combined.

LEVEL OF EVIDENCE: Diagnostic study, Level III.

PMID:40293511 | DOI:10.1007/s00264-025-06531-2

Sleep. 2025 Apr 28:zsaf113. doi: 10.1093/sleep/zsaf113. Online ahead of print.

ABSTRACT

Wearable devices with sleep tracking functionalities can prompt behavioral changes to promote sleep, but proactively preventing poor sleep when it is likely to occur remains a challenge due to a lack of prediction models that can forecast sleep parameters prior to sleep onset. We developed models that forecast low sleep efficiency 4 and 8 hours prior to sleep onset using gradient boosting (CatBoost) and deep learning (Convolutional Neural Network Long Short-Term Memory, CNN-LSTM) algorithms trained exclusively on accelerometer data from 80,811 adults in the UK Biobank. Associations of various sleep and activity parameters with sleep efficiency were further examined. During repeated cross-validation, both CatBoost and CNN-LSTM exhibited excellent predictive performance (median AUCs > 0.90, median AUPRCs > 0.79). U-shaped relationships were observed between total activity within 4 and 8 hours of sleep onset and low sleep efficiency. Functional data analyses revealed higher activity 6 to 8 hours prior to sleep onset had negligible associations with sleep efficiency. Higher activity 4 to 6 hours prior had moderate beneficial associations, while higher activity within 4 hours had detrimental associations. Additional analyses showed that increased variability in sleep duration, efficiency, onset timing, and offset timing over the preceding four days was associated with lower sleep efficiency. Our study represents a first step towards wearable-based machine learning systems that proactively prevent poor sleep by demonstrating that sleep efficiency can be accurately forecasted prior to bedtime and by identifying pre-bed activity targets for subsequent intervention.

PMID:40293116 | DOI:10.1093/sleep/zsaf113

Sensors (Basel). 2025 Mar 19;25(6):1924. doi: 10.3390/s25061924.

ABSTRACT

With the development of UAV technology, the composition of UAVs has become increasingly complex, interconnected, and tightly coupled. Fault features are characterized by weakness, nonlinearity, coupling, and uncertainty. A promising approach is the use of deep learning methods, which can effectively extract useful diagnostic information from weak, coupled, nonlinear data from inputs with background noise. However, due to the diversity of flight environments and missions, the distribution of the obtained sample data varies. The types of fault data and corresponding labels under different conditions are unknown, and it is time-consuming and expensive to label sample data. These challenges reduce the performance of traditional deep learning models in anomaly detection. To overcome these challenges, a novel weighted-transfer domain-adaptation network (WTDAN) method is introduced to realize the online anomaly detection and fault diagnosis of UAV electromagnetic-sensitive flight data. The method is based on unsupervised transfer learning, which can transfer the knowledge learnt from existing datasets to solve problems in the target domain. The method contains three novel multiscale modules: a feature extractor, used to extract multidimensional features from the input; a domain discriminator, used to improve the imbalance of the data distribution between the source domain and the target domain; and a label classifier, used to classify data categories for the target domain. Multilayer domain adaptation is used to reduce the distance between the source domain datasets and the target domain datasets distributions. The WTDAN assigns different weights to the source domain samples in order to weight the different contributions of source samples to solve the problem during the training process. The dataset adopts not only open datasets from the website but also test datasets from experiments to evaluate the transferability of the proposed WTDAN model. The experimental results show that, under the condition of fewer anomalous target data samples, the proposed method had a classification accuracy of up to 90%, which is higher than that of the other compared methods, and performed with superior transferability on the cross-domain datasets. The capability of fault diagnosis can provide a novel method for online anomaly detection and the prognostics and health management (PHM) of UAVs, which, in turn, would improve the reliability, repairability, and safety of UAV systems.

PMID:40293102 | DOI:10.3390/s25061924

Sensors (Basel). 2025 Mar 20;25(6):1942. doi: 10.3390/s25061942.

ABSTRACT

Fringe projection profilometry (FPP) is a widely employed technique owing to its rapid speed and high accuracy. However, when FPP is utilized to measure shiny surfaces, the fringes tend to be saturated or too dark, which significantly compromises the accuracy of the 3D measurement. To overcome this challenge, this paper proposes an efficient method for the 3D measurement of shiny surfaces based on FPP. Firstly, polarizers are employed to alleviate fringe saturation by leveraging the polarization property of specular reflection. Although polarizers reduce fringe intensity, a deep learning method is utilized to enhance the quality of fringes, especially in low-contrast regions, thereby improving measurement accuracy. Furthermore, to accelerate measurement efficiency, a dual-frequency complementary decoding method is introduced, requiring only two auxiliary fringes for accurate fringe order determination, thereby achieving high-efficiency and high-dynamic-range 3D measurement. The effectiveness and feasibility of the proposed method are validated through a series of experimental results.

PMID:40293081 | DOI:10.3390/s25061942

Sensors (Basel). 2025 Mar 20;25(6):1926. doi: 10.3390/s25061926.

ABSTRACT

This paper presents a new methodology for localizing impact events on plate-like structures using a proposed two-dimensional convolutional neural network (CNN) and received impact signals. A network of four piezoelectric wafer active sensors (PWAS) was installed on the tested plate to acquire impact signals. These signals consisted of reflection waves that provided valuable information about impact events. In this methodology, each of the received signals was divided into several equal segments. Then, a wavelet transform (WT)-based time-frequency analysis was used for processing each segment signal. The generated WT diagrams of these segments' signals were cropped and resized using MATLAB code to be used as input image datasets to train, validate, and test the proposed CNN model. Two scenarios were adopted from PAWS transducers. First, two sensors were positioned in two corners of the plate, while, in the second scenario, four sensors were used to monitor and collect the signals. Eight datasets were collected and reshaped from these two scenarios. These datasets presented the signals of two, three, four, and five impacts. The model's performance was evaluated using four metrics: confusion matrix, accuracy, precision, and F1 score. The proposed model demonstrated exceptional performance by accurately localizing all of the impact points of the first scenario and 99% of the second scenario. The main limitation of the proposed model is how to differentiate the data samples that have similar features. From our point of view, the similarity challenge arose from two factors: the segmentation interval and the impact distance. First, applying the segmenting procedure to the PWAS signals led to an increase in the number of data samples. The procedure segmented each PWAS signal to 30 samples with equal intervals, regardless of the features of the signal. Segmenting and transforming different PWAS signals into image-based data points led to data samples that had similar features. Second, some of the impacts had a close distance to the PWAS sensors, which resulted in similar segmented signals. Therefore, the second scenario was more challenging for the proposed model.

PMID:40293079 | DOI:10.3390/s25061926

Sensors (Basel). 2025 Mar 19;25(6):1902. doi: 10.3390/s25061902.

ABSTRACT

Advanced communication systems, particularly in the context of autonomous driving and integrated sensing and communication (ISAC), require high precision and refresh rates for environmental perception, alongside reliable data transmission. This paper presents a novel approach to enhance the ISAC performance in existing 4G and 5G systems by utilizing a two-dimensional offset in the Delay-Doppler (DD) domain, effectively leveraging the sounding reference signal (SRS) resources. This method aims to improve spectrum efficiency and sensing accuracy in vehicular networks. However, a key challenge arises from interference between multiple users after the wireless propagation of signals. To address this, we propose a deep learning-based interference mitigation solution using an UNet architecture, which operates on the Range-Doppler maps. The UNet model, with its encoder-decoder structure, efficiently filters out unwanted signals, therefore enhancing the system performance. Simulation results show that the proposed method significantly improves the accuracy of environmental sensing and resource utilization while mitigating interference, even in dense network scenarios. Our findings suggest that this DD-domain-based approach offers a promising solution to optimizing ISAC capabilities in current and future communication systems.

PMID:40293069 | DOI:10.3390/s25061902

Sensors (Basel). 2025 Mar 19;25(6):1909. doi: 10.3390/s25061909.

ABSTRACT

Echocardiography is a widely used cardiac imaging modality in clinical practice. Physicians utilize echocardiography images to measure left ventricular volumes at end-diastole (ED) and end-systole (ES) frames, which are pivotal for calculating the ejection fraction and thus quantitatively assessing cardiac function. However, most existing approaches focus on features from ES frames and ED frames, neglecting the inter-frame correlations in unlabeled frames. Our model is based on an encoder-decoder architecture and consists of two modules: the Temporal Feature Fusion Module (TFFA) and the Vision Retentive Network (Vision RetNet) encoder. The TFFA leverages self-attention to learn inter-frame correlations across multiple consecutive frames and aggregates the features of the temporal-channel dimension through channel aggregation to highlight ambiguity regions. The Vision RetNet encoder introduces explicit spatial priors by constructing a spatial decay matrix using the Manhattan distance. We conducted experiments on the EchoNet-Dynamic dataset and the CAMUS dataset, where our proposed model demonstrates competitive performance. The experimental results indicate that spatial prior information and inter-frame correlations in echocardiography images can enhance the accuracy of semantic segmentation, and inter-frame correlations become even more effective when spatial priors are provided.

PMID:40293054 | DOI:10.3390/s25061909